Evaluation of Proteoglycan Gene Polymorphisms as Risk Factors in the Genetic Susceptibility to High Myopia

Abstract

Purpose.: To investigate the relationship between high myopia and single nucleotide polymorphisms (SNPs) in six proteoglycan genes: aggrecan (ACAN), fibromodulin (FMOD), decorin (DCN), lumican (LUM), keratocan (KERA), and epiphycan (EPYC). These genes were selected for study because they are involved in induced myopia in animals and/or are within the human MYP3 locus identified by linkage analysis of families with high myopia.

Methods.: Two groups of Chinese subjects were studied: group 1 (300 cases and 300 controls) and group 2 (356 cases and 354 controls). Cases were high myopes with spherical equivalent (SE) ≤ −8.00 D, and controls had SE between +1.0 and −1.0 D. From these candidate genes, 60 tagging SNPs were selected. First, 12 DNA pools were each constructed from 50 samples of the same phenotype from group 1 subjects and were tested for association with the SNPs. Second, putatively positive SNPs were confirmed by individual genotyping of group 1 subjects. Finally, positive results were replicated in group 2 subjects.

Results.: Of the 58 SNPs successfully screened by DNA pooling, 8 ACAN SNPs passed the threshold of P ≤ 0.10 (nested ANOVA) and were then genotyped in the individual samples. Haplotypes rs3784757 and rs1516794 showed significant association with high myopia. However, the positive result could not be replicated in the second subject group.

Conclusions.: These six proteoglycan genes were not associated with high myopia in these Chinese subjects and hence are unlikely to be important in the genetic predisposition to high myopia.

Myopic eyes focus distant objects in front of, instead of on, the retina. Myopia is the most frequent ocular disorder worldwide, with a wide range of prevalence in different populations. It is much more frequent in Asian populations (50%–70%) than in Caucasians populations (up to 30%). ¹ High myopia, usually defined as −5.00 D or worse, is a predisposing factor for many pathologic ocular complications such as glaucoma and retinal detachment. ¹ It is a multifactorial disease caused by genetic factors, environmental factors, and their interactions. ¹  

Myopia usually develops as a result of excessive elongation of the eyeball with concomitant scleral remodelling that involves changes in the metabolism of collagen fibrils and proteoglycans. ^2,3 Proteoglycans are proteins that are heavily glycosylated and are each composed of a core protein covalently linked with at least one glycosaminoglycan chain. They are important in regulating the assembly and interaction of collagen fibrils and scleral hydration although they contribute less than 1% to the dry weight of the sclera. ^2,3 Proteoglycans found in the sclera include aggrecan (ACAN), fibromodulin (FMOD), lumican (LUM), decorin (DCN), biglycan (BGN), keratocan (KERA), epiphycan (EPYC) and others. ^2,3  

In chicks with myopia induced by form deprivation, a proteoglycan identified as aggrecan was found to accumulate in increased amounts in the presence of increased turnover rate in the cartilaginous layer of the posterior sclera, in parallel with the scleral growth of the treated eye, when compared with the control eye. ^4,5 Interestingly, the changes were in the opposite direction in the fibrous scleral layer of the chick and the sclera (fibrous in nature) of mammals such as the tree shrew and monkey. ⁶ In the tree shrew, synthesis of glycosaminoglycan (and hence proteoglycan) was decreased in the posterior sclera of form-deprived eyes, with accompanying axial elongation and scleral thinning, when compared with the control eye. ⁷ This effect could be explained by reduced ACAN gene expression, as shown in the sclera of lens-induced myopic eyes in the tree shrew. ⁸ In induced myopia, alteration of DCN synthesis has been demonstrated in the sclera of the elongated eye in the chick ⁴ and the marmoset, ⁹ but not in the tree shrew. ^8,10 BGN and LUM showed little differential regulation in the sclera in response to lens-induced myopia in the tree shrew. ⁸ Intriguingly, features of high myopia (thin sclera, increased axial length, and retinal detachment) were found in the eyes of lumican-fibromodulin double-null mice. ¹¹ Linkage analysis of high myopia families has identified the MYP3 locus at chromosome 12, region q21-23, and the proteoglycan genes DCN, LUM, KERA, and EPYC lie adjacent to each other within this locus. ¹²  

This biological and positional information justifies our investigation of these proteoglycan genes as candidate genes for high myopia. An efficient stepwise DNA pooling-based case–control study approach ^13,14 was used to investigate whether common tagging single nucleotide polymorphisms (tSNPs) of these genes were associated with high myopia in a Chinese subject sample. We examined six proteoglycan genes: ACAN, FMOD, DCN, LUM, KERA, and EPYC (Table 1). BGN is an X-linked gene and hence cannot be studied by a DNA pooling-based approach in which DNA samples from male and female subjects are randomly mixed to construct DNA pools. 

Table 1.

 

Features of Candidate Genes and Their tSNPs

Gene	Gene Name	Gene ID	Chromosome Location	Region Captured*	tSNPs n	SNPs Captured by tSNPs n (r 2)
ACAN	Aggrecan	176	15q26.1	76.8 kb	41	106 (0.964)
FMOD	Fibromodulin	2331	1q32	16.5 kb	9	23 (0.955)
DCN	Decorin	1634	12q21.33	43.8 kb	2	3 (1.000)
LUM	Lumican	4060	12q21.3-22	14.3 kb	3	15 (0.982)
KERA	Keratocan	11081	12q22	13.9 kb	2	17 (1.000)
EPYC	Epiphycan	1833	12q21	47.3 kb	3	31 (1.000)

*  The region captured includes the gene and its 6-kb adjoining region (3 kb upstream and 3 kb downstream).

Materials and Methods

Overview of the Study Approach

The first stage was a screen of separate case and control DNA pools to discover putatively positive SNPs. The second stage was to confirm these putatively positive SNPs by genotyping individual DNA samples that formed the original DNA pools. The third stage was to replicate the confirmed positive SNPs with a new sample set. 

Study Subjects

For the first and the second stage of the study, we recruited 600 unrelated ethnic Chinese (group 1 subjects): 300 with high myopia and 300 emmetropic controls. ¹⁵ For the third stage of the study, we recruited 710 unrelated ethnic Chinese (group 2 subjects): 356 with high myopia and 354 emmetropic controls. ¹⁶ We used the same recruitment criteria for both subject groups: High myopia was defined as spherical equivalent (SE) ≤ −8.00 D in both eyes and emmetropia as SE between +1.0 and −1.0 D in both eyes. The characteristics of the subjects have been reported ^15,16 and are summarized here. In the cases, the average SE and axial length were −10.53 D and 27.76 mm in group 1 and −10.30 D and 27.64 mm in group 2, respectively. For controls, the mean SE and axial length were 0.03 D and 23.85 mm in group 1 and 0.08 D and 23.73 mm in group 2, respectively. Group 1 subjects were younger than group 2 subjects (26.1 years vs 33.6 years). 

Ethics approval was obtained from the Human Subjects Ethics Subcommittee of the Hong Kong Polytechnic University, and the protocol complied with the Declaration of Helsinki. We recruited all subjects with written informed consent in the Optometry Clinic of our University, collected blood samples, and extracted DNA from blood samples as has been reported. ¹⁵  

DNA Pool Construction

DNA concentration was accurately measured (PicoGreen Kit; Invitrogen, Carlsbad, CA) for DNA samples (group 1 subjects). DNA samples at 5.0 ± 0.3 ng/μL were mixed in equal amounts to create DNA pools. DNA pools were each created from 50 different subjects of the same disease status: 6 case pools from 300 case samples and 6 control pools from 300 control samples. 

Genotyping of DNA Pools and Individual Samples

From the six candidate genes ACAN, FMOD, DCN, LUM, KERA, and EPYC, 60 tSNPs were selected using Tagger software (http://www.broadinstitute.org/mpg/tagger/, Broad Institute, Cambridge, MA) with the criteria of pairwise tagging, a correlation coefficient (r ²) of at least 0.8 and a minor allele frequency (MAF) of at least 0.1, and based on HapMap Han Chinese data (release 23a, phase II; http://www.hapmap.org/) (Table 1). For each of the selected tSNPs, DNA pools were amplified using touchdown polymerase chain reaction (PCR), as described previously ¹⁶ with specific primers and conditions shown in Supplementary Table S1. The PCR products were purified and then used as templates for primer extension (PE). ¹⁶ The PE products were injected into a denaturing high-performance liquid chromatography (DHPLC) system (WAVE Nucleic Acid Fragment Analysis System; Transgenomic, Omaha, NE) to estimate the relative allele frequencies of the two alleles for each SNP. ¹⁶ Each DNA pool was independently tested and analyzed three times for each SNP, and hence each SNP had a total of 36 sets of readings from 12 DNA pools. Differential incorporation of dideoxynucleotides in PE was corrected by means of a correction factor (k), ¹⁷ which was obtained as the mean value of three independent analyses of a heterozygous sample for each SNP. 

In the second stage of the study, individual samples of group 1 subjects were genotyped by mass spectrometry of multiplexed PE products (MassArray iPLEX assays; Sequenom, San Diego, CA), according to the recommended protocols (http://www.sequenom.com/); primer sequences are shown in Supplementary Table S2. Putatively positive SNPs from the first stage were grouped together with genetic markers of other studies and genotyped using this method by a local genotyping laboratory (Hong Kong University; http://genome.hku.hk/portal/). One SNP (rs1516794) could not be grouped with other SNPs for genotyping by the commercial platform, and was genotyped in-house by restriction analysis. In the third stage, two confirmed SNPs (rs3784757 and rs1516794) were further genotyped for individual samples from group 2 subjects by in-house methods based on restriction analysis (Supplementary Table S2). 

Statistical Analysis

The relative allele frequency data from DNA pools were analyzed (STATA ver. 8.2; StataCorp, College Station, TX). For each SNP, the relative allele frequencies were calculated from the peak intensities of the two PE products with adjustment based on the k correction factor as described previously. ¹⁷ Nested analysis of variance (ANOVA) was used to compare the relative allele frequencies obtained from the case pools and the control pools because DNA pools were nested separately within the cases and the controls and there was no link between any case pools to any control pools. ¹⁶ To avoid excluding potential significant SNP, we used a P value ≤0.10 as the cutoff for follow-up putatively positive SNPs in the second stage. 

Linkage disequilibrium (LD) measures were calculated and plotted using Haploveiw (version 4.2; http://www.broad.mit.edu/mpg/haploview/). Data of individual genotypes (second and third stages) were analyzed by Plink (ver. 1.07; http://pngu.mgh.harvard.edu/∼purcell/plink/index.shtml). An exact test was used to test for Hardy-Weinberg equilibrium (HWE) ¹⁸ in cases and controls separately. Logistic regression was used for testing association between high myopia and single markers and their haplotypes. Exhaustive haplotype analysis was performed with a sliding-window strategy for windows of all possible sizes (i.e., one SNP to seven SNPs per window). To avoid potential confounding, both sex and age were included as covariates for adjustment in all analyses. Wald test gave an asymptotic P value (P _asym) for each test. To correct for multiple comparisons, we used a permutation test to permute the phenotype status of the subjects without changing the genotypes across all single markers and all haplotypes for samples individually genotyped within a subject group (group 1, group 2, or combined groups, each separately). We generated empiric P values (P _emp) based on 10,000 permutations within each subject group. 

Results

Stage 1: Analysis of DNA Pools from Group 1 Subjects

We failed to find a heterozygous sample after screening 40 samples for two SNPs (rs7174219 and rs7173022) of the ACAN gene, which were then dropped from testing. Thus, 58 were successfully analyzed by the DNA-pooling approach (Table 2). The mean k correction factor was 0.9488 (range, 0.5645–2.6026). The first eluted allele had an estimated frequency ranging from 0.0811 to 0.8991 in the cases and from 0.1130 to 0.9037 in the controls. Estimated differences in allele frequencies (case pools minus control pools) ranged from −0.0548 to 0.0434. Of the 58 tSNPs tested, 8 showed a significant difference in allele frequencies at a cutoff of P ≤ 0.10, all belonging to the ACAN gene (Table 2). For confirmation, these 8 SNPs were genotyped for individual samples that formed the original DNA pools (group 1 subjects). No significant difference in allele frequencies was detected for the remaining 50 SNPs, which were thus not tested any further. 

Table 2.

 

Pooled DNA Analysis of the tSNPs

Candidate Gene	SNP*	Alleles† (1st/2nd)	k Correction Factor Peak Height Ratio (1st/2nd)	Estimated Frequency of 1st Allele in DNA Pools			Nested ANOVA P
				Case	Control	Diff (Case − Control)
ACAN	rs2203642	G/A	0.8991	0.4179	0.4407	−0.0228	0.3202
	rs12439075‡	C/T	1.0880	0.4051	0.4461	−0.0410	0.0862
	rs8033375	C/T	0.6940	0.3178	0.3290	−0.0112	0.5053
	rs4932429	G/C	1.1401	0.8991	0.8689	0.0302	0.1916
	rs11858871	G/T	0.9799	0.1593	0.1946	−0.0353	0.1479
	rs17199220	G/A	0.9069	0.8331	0.8122	0.0209	0.5378
	rs16942248	T/A	1.0196	0.1721	0.1714	0.0007	0.9739
	rs12905259	G/A	0.9024	0.5185	0.4965	0.0220	0.2407
	rs12905452	A/C	1.0506	0.5257	0.5448	−0.0191	0.3009
	rs7179602	T/A	2.6026	0.2653	0.2769	−0.0116	0.4223
	rs4932433	A/T	1.1017	0.3360	0.3302	0.0058	0.8087
	rs16942277	G/T	0.9047	0.7090	0.7004	0.0086	0.6283
	rs4932434	C/A	0.9939	0.5989	0.5979	0.0010	0.9605
	rs939586	G/A	0.9888	0.2369	0.2399	−0.0030	0.9134
	rs11073814	A/T	1.0536	0.8451	0.8442	0.0009	0.9688
	rs8040435	T/A	1.0101	0.5917	0.5908	0.0009	0.9722
	rs2280468‡	G/A	0.9654	0.7338	0.6974	0.0364	0.0762
	rs1015081	G/A	0.8835	0.1868	0.1748	0.0120	0.6409
	rs1015080	G/A	0.8955	0.1252	0.1521	−0.0269	0.1928
	rs883325	G/T	1.0082	0.2962	0.2804	0.0158	0.4727
	rs4932435	G/T	0.9796	0.5223	0.5080	0.0143	0.5982
	rs2293087‡	G/T	0.8570	0.3109	0.2737	0.0372	0.0338
	rs4932438‡	C/T	1.1072	0.2270	0.2818	−0.0548	0.0142
	rs3743398	C/T	0.8793	0.8990	0.9037	−0.0047	0.7613
	rs938608	G/T	0.8455	0.6948	0.7125	−0.0177	0.4717
	rs4932439	G/A	0.6748	0.4747	0.4896	−0.0149	0.5029
	rs1042631	C/T	0.9011	0.6006	0.6211	−0.0205	0.4229
	rs698621	G/T	0.9484	0.4812	0.4461	0.0351	0.1423
	rs953065‡	C/T	0.8001	0.4959	0.5358	−0.0399	0.0785
	rs3784757‡	G/A	0.9776	0.8457	0.8201	0.0256	0.0346
	rs1516793‡	G/A	0.9921	0.1391	0.1606	−0.0215	0.0982
	rs1516794‡	T/A	1.0360	0.0811	0.1130	−0.0319	0.0542
	rs1516797	G/T	0.9280	0.329	0.3207	0.0083	0.6185
	rs1879529	G/T	0.9867	0.7336	0.7140	0.0196	0.2597
	rs3817428	C/G	1.0224	0.8572	0.8469	0.0103	0.5798
	rs16942409	G/T	0.8121	0.2824	0.2786	0.0038	0.8381
	rs7163146	A/T	0.5858	0.1443	0.1576	−0.0133	0.6233
	rs2280465	G/A	0.7680	0.8701	0.8730	−0.0029	0.8892
	rs8031741	G/A	0.9224	0.1496	0.1513	−0.0017	0.9627
FMOD	rs10920617	C/T	0.8611	0.7186	0.6885	0.0301	0.1331
	rs10920615	C/T	0.6669	0.5774	0.5628	0.0146	0.2744
	rs7543148	G/A	0.9589	0.3008	0.3225	−0.0217	0.3377
	rs10800913	C/T	0.8889	0.2549	0.2591	−0.0042	0.8300
	rs2105309	C/T	0.9142	0.6212	0.6244	−0.0032	0.8281
	rs3766913	G/A	0.9426	0.8716	0.8773	−0.0057	0.6797
	rs3820224	G/A	0.9440	0.8088	0.7654	0.0434	0.1191
	rs2886220	G/A	0.8671	0.4865	0.4965	−0.0100	0.6271
	rs16851319	C/G	0.7840	0.2402	0.1986	0.0416	0.1072
DCN	rs3138295	G/A	0.6552	0.2696	0.2428	0.0268	0.4038
	rs566806	C/T	0.9447	0.6043	0.6039	0.0004	0.9851
LUM	rs3759222	C/A	0.9089	0.6220	0.6486	−0.0266	0.4192
	rs10859110	C/T	0.9339	0.5868	0.5793	0.0075	0.7974
	rs2300588	G/T	0.7815	0.3538	0.3345	0.0193	0.4250
KERA	rs2041711	G/T	0.9056	0.8214	0.8109	0.0105	0.6897
	rs2268579	G/A	0.5645	0.1430	0.1324	0.0106	0.6827
EPYC	rs11105899	T/A	1.1288	0.8827	0.8558	0.0269	0.5534
	rs11105898	C/T	1.1958	0.4723	0.4970	−0.0247	0.5943
	rs10859081	G/A	1.0734	0.1850	0.1826	0.0024	0.9200

*  SNPs are arranged down the column in the 5′→3′ order along the sense strand of the gene concerned.

†  The 1st allele is eluted first (shorter elution), and the 2nd allele is eluted last (longer elution time). The alleles are named with reference to the sense strand of the respective gene.

‡  SNPs with P ≤ 0.1 are highlighted in bold and are followed up by individual genotyping in the second stage of the study.

Stage 2: Confirming Pooled DNA Results from Group 1 Subjects by Individual Genotyping

One genotyped SNP (rs1516793) (MassArray iPLEX; Sequenom) failed to pass filtering quality checks because of poor assay performance and was thus not included in subsequent data analysis. The remaining seven SNPs were designated as S1 to S7 for easy referencing (Fig. 1; Tables 3, 4). The genotypes of these seven SNPs were in HWE (P > 0.05) in the group 1 subjects. The LD among the SNPs was in general very weak (Fig. 2) although two LD blocks could be constructed. Single-marker analysis did not reveal any significant difference in allele frequencies between cases and controls (P _emp >0.05, Table 3). However, exhaustive sliding-window–based haplotype analysis identified a 2-SNP window that showed a significant difference in haplotype frequencies between cases and controls (P _asym = 0.0002 and P _emp = 0.0017 for S6…S7 or rs3784757 and rs1516794; Tables 4 and 5). Of all 28 possible sliding windows, 16 gave P _asym ≤ 0.05, but only the S6…S7 window showed association with high myopia after correction for multiple comparisons (Table 4). Therefore, these two SNPs were further tested in the third stage of the study (replication study), using group 2 subjects. The remaining five SNPs were dropped from further examination. 

Figure 1.

 

The structure of the aggrecan (ACAN) gene and the seven SNPs tested in the second stage of the study. Top: alternative exon–intron organizations of the ACAN gene taken from the UCSC Genome Browser (http://genome.ucsc.edu/cgi-bin/hgGateway, University of California at Santa Cruz) together with their corresponding positions on chromosome 15 based on the GRCh37/hg19 human reference sequence assembly. Bottom: the physical positions of the seven intronic SNPs tested in the second stage of the study.

Table 3.

 

Allelic Association Tests of ACAN SNPs Genotyped Individually

SNP	Alleles*		Genotype Counts (11/12/22)*		Minor Allele Frequency		OR (95% CI)†	Allelic Test‡
	1	2	Cases	Controls	Cases	Controls		P asym	P emp
Group 1 Subjects
rs12439075 (S1)	T	C	118/134/44	106/140/52	0.3750	0.4094	0.87 (0.69–1.11)	0.2667	0.9319
rs2280468 (S2)	G	A	162/104/14	143/123/15	0.2357	0.2823	0.78 (0.58–1.04)	0.0879	0.5690
rs2293087 (S3)	T	G	147/130/21	170/117/10	0.2886	0.2306	1.40 (1.05–1.86)	0.0224	0.1936
rs4932438 (S4)	T	C	198/83/16	172/103/22	0.1936	0.2475	0.71 (0.53–0.93)	0.0148	0.1333
rs953065 (S5)	C	T	79/141/68	89/138/64	0.4809	0.4570	1.11 (0.88–1.41)	0.3732	0.9801
rs3784757 (S6)	G	A	243/53/3	232/62/5	0.0987	0.1204	0.74 (0.51–1.08)	0.1234	0.6957
rs1516794 (S7)	A	T	246/34/0	241/47/3	0.0607	0.0911	0.60 (0.37–0.96)	0.0326	0.2717
Group 2 Subjects
rs3784757 (S6)	G	A	281/70/3	282/67/4	0.1073	0.1062	1.02 (0.72–1.43)	0.9213	0.9916
rs1516794 (S7)	A	T	304/50/0	302/47/1	0.0706	0.0700	1.02 (0.67–1.55)	0.9373	0.9960
Combined (Groups 1 and 2)
rs3784757 (S6)	G	A	524/123/6	514/129/9	0.1034	0.1127	0.90 (0.70–1.15)	0.3998	0.6024
rs1516794 (S7)	A	T	550/84/0	543/94/4	0.0662	0.0796	0.82 (0.60–1.11)	0.1976	0.3389

*  Alleles 1 and 2 refer to the major and the minor alleles, respectively,; and the genotype counts are indicated in the order of 11, 12, and 22. There were 300 cases and 300 controls in group 1, and 356 cases and 354 controls in group 2. Since some samples failed to be genotyped in a random manner, the total counts of genotypes may not add up to the expected numbers.

†  The odds ratio (OR) is calculated for allele 2 (minor allele) with allele 1 (major allele) as the reference, with 95% CI.

‡  Allele frequencies are compared by logistic regression with adjustment for sex and age as covariates to give the asymptotic P values (P _asym). Multiple comparisons are corrected by 10,000 permutations across single marker allelic tests (this table) and omnibus tests of haplotypes in group 1 subjects (Table 4), in group 2 subjects (Table 5) or combined (groups 1 and 2, Table 5), each separately.

Table 4.

 

Summary of Exhaustive Haplotype Analyses Based on Sex- and Age-Adjusted Omnibus Tests for Sliding Windows of All Possible Sizes across Seven ACAN SNPs Genotyped Individually for Group 1 Subjects*

Sliding Window		Most Significant Omnibus Test
SNPs/SW	No. of SW	SW	P asym	P emp
1	7	S4	0.0148	0.1333
2	6	S6 … S7	0.0002	0.0017
3	5	S4 … S6	0.0078	0.0753
4	4	S3 … S6	0.0088	0.0829
5	3	S3 … S7	0.0020	0.1782
6	2	S2 … S7	0.0156	0.1400
7	1	S1 … S7	0.0152	0.1365

*  Plink was used to compare haplotypes between cases and controls in all sliding windows (SW) of all possible window sizes (the number of SNPs per SW; 1 to 7 SNPs per SW). With logistic regression and adjustment for sex and age as covariates, a single omnibus test of (n − 1) degrees was performed for each SW, to jointly assess the significance of the haplotype effects for this SW, where n is the number of haplotypes for the specific SW being considered. Hence, a single asymptotic P value (P _asym) was generated for each SW. For a given window size, the omnibus test was conducted for all possible windows of the same size, shifting one SNP at a time. There were 28 windows in total (the sum of numbers in the second column from the left), and multiple comparisons were corrected by performing 10,000 permutations to produce an empirical P value (P _emp). The SW is shown as Sx … Sy, where x is the first SNP and y the last SNP of the SW, and the SW S3-S4-S5-S6, for instance, is indicated as S3 … S6. The rightmost columns show the most significant omnibus tests for each fixed-size SW. Note that only S6 … S7 remained significant after multiple testing correction (P _emp = 0.0017; indicated in bold). Please refer to Table 3 for the identities of the SNPs S1 to S7.

Figure 2.

 

The LD pattern of seven SNPs of the ACAN gene. The SNPs are indicated in the 5′→3′ direction (left→right) along the sense strand of the gene. Shown are the LD measures expressed as D′ and r ² for all subjects of group 1 and calculated by Haploview. The shades of gray represent the magnitude of the LD measures.

Table 5.

 

Haplotype Analysis of Two ACAN SNPs: rs3784757 and rs1516794*

Haplotype	Haplotype Frequency		Logistic Regression Adjusted for Sex and Age
	Cases	Controls	OR	P asym	P emp
Group 1 Subjects
Omnibus test	—	—	—	0.0002	0.0017
AT	0.0373	0.0871	0.37	0.0005
GT	0.0237	0.0026	12.50	0.0119
AA	0.0595	0.0370	1.61	0.1150
GA	0.8795	0.8733	1.13	0.5100
Group 2 Subjects
Omnibus test	—	—	—	0.9940	1.0000
AT	0.0697	0.0689	1.02	0.9280
AA	0.0370	0.0373	0.98	0.9550
GA	0.8933	0.8938	0.99	0.9660
Combined (Groups 1 and 2)
Omnibus test	—	—	—	0.2670	0.4369
AT	0.0561	0.0776	0.69	0.0238
AA	0.0473	0.0368	1.26	0.2510
GA	0.8966	0.8856	1.03	0.8020

*  Asymptotic P values (P _asym) are obtained from the Wald test, based on logistic regression. Multiple comparisons are corrected by 10,000 permutations across single-marker allelic tests (Table 3) and omnibus tests of haplotypes (this table) for group 2 subjects or combined groups, each separately. For group 1 subjects, the result was an expansion of the sliding window S6 … S7 shown in Table 4, and the correction for multiple comparisons was performed for all single-marker allelic tests (Table 3) and haplotype tests of all sliding windows (Table 4). The empirical P is indicated as P _emp. Note that Plink does not generate confidence intervals for odds ratios (OR) in haplotype analysis.

Stage 3: Replication Study Using Group 2 Subjects

For rs3784757 and rs1516794, the genotypes of group 2 subjects were in HWE (P > 0.05). No significant difference in allele frequencies (Table 3) and haplotype frequencies (Table 5) was revealed between cases and controls. Therefore, the initial positive results obtained with group 1 subjects were not substantiated in a second group of subjects of the same ethnicity. Further analysis was performed by combining both subject groups (656 cases and 654 controls in total). Significant difference between cases and controls in frequencies of alleles and haplotypes were still not detected (Tables 3, 5). 

Discussion

Six proteoglycan genes ACAN, FMOD, DCN, LUM, KERA, and EPYC were selected for study because of their involvement in induced myopia in animal models and/or being within the MYP3 interval identified by linkage analysis of families with high myopia. ^{4 –12} In particular, ACAN and FMOD were selected as biological candidate genes, KERA and EPYC as positional candidate genes, and DCN and LUM as biological and positional candidate genes. Association studies of candidate genes selected on the basis of biological and/or positional information have meet with different levels of success. However, there are indeed examples of myopia susceptibility genes identified from each approach: TGFB1 from a biology-based approach ^17,19,20 ; MFN1, SOX2OT and PSARL from a position-based approach (MYP8 locus) ²¹ ; and HGF and PAX6 from a combined approach based on both biological and positional information. ^{22 –27}  

Sixty tSNPs were selected from the six selected genes (Table 1), 58 tSNPs were successfully screened by DNA pooling strategy (Table 2), and 8 tSNPs (P ≤ 0.1, nested ANOVA, Table 2) were followed up by individual genotyping of 300 cases and 300 controls. Although single-marker analysis did not reveal any significant results (Table 3), a sliding-window–based haplotype analysis identified significant difference between cases and controls in the frequencies of haplotypes consisting of rs3784757 and rs1516794 (S6…S7, Table 4). However, the attempt to validate these findings in a second subject group (356 cases and 354 controls) failed to replicate the initial positive results. In other words, these six proteoglycan genes were not associated with high myopia in the Chinese population under study and are therefore unlikely to make a major contribution to the genetic predisposition to high myopia. 

To exclude type II error as a possible explanation for the lack of positive association results, we examine the power of our study. To calculate the power of the first stage of the study (screening tSNPs by DNA pooling approach), an online calculator for power and sample size (http://www.stat.uiowa.edu/∼rlenth/Power/, University of Iowa) was used. In our nested ANOVA model,¹⁶ the subject group was a fixed-effects factor with two levels (case and control) while the DNA pool was a random-effects factor with six levels (six DNA pools per subject group). The technical replicate measurement also assumed a random effect. For random effects and from the analysis output of the statistical package (STATA; StataCorp.), the effect size expressed as the square root of the variance component was on average 0.0356 for the factor DNA pool and 0.0099 for the technical replicate measurements. An allele frequency difference of 0.015 (1.5%) between the subject groups was translated into an effect size of 0.0441 for the fixed-effects factor subject group and calculated as the square root of the sum of squares (from STATA analysis output) divided by the degree of freedom (df = 1). With the significance level set at α = 0.10 for this screening stage, an SNP showing an allele frequency difference of 1.5% between the subject groups was detected with a power of 89% (last row of Supplementary Table S3 and hence would be followed up in the second stage by individual genotyping. We tested seven SNPs in the second stage of the study. We assume a prevalence of 0.05 for high myopia in the general Chinese population of Hong Kong²⁸ and a Bonferroni-adjusted significance of 0.0071 for 7 SNPs, which is much more conservative than the permutation tests used in the data analysis. Under a log-additive model, as examined using the QUANTO package (ver. 1.2.4),²⁹ a sample size of 300 cases and 300 controls would give a power of ∼80%, with an odds ratio of 1.85 and a minor allele frequency of 0.10 (Supplementary Table S4). We tested two SNPs in the third stage of the study. With similar assumptions and a Bonferroni-adjusted significance of 0.025 for two SNPs, a sample size of 356 cases and 356 controls would give a power of ∼80%, with an odds ratio of 1.65 and a minor allele frequency of 0.10 (Supplementary Table S4). In other words, the power is at least 80% for all parts of the study under reasonable assumptions and using the empiric data from the study. 

Although the role of ACAN and KERA common polymorphisms in high myopia was investigated for the first time in this study, the other four genes have been examined previously in relation to high myopia in smaller studies. One group examined one FMOD SNP (rs7543418), but did not find any association in a study involving 195 Chinese cases (SE ≤ −6.5 D) and 94 controls. ³⁰ With a DNA pooling approach, we examined nine tSNPs from FMOD and did not find any association for rs7543418 either (Table 2). Another group explored four DCN and four EPYC polymorphisms, with only four (rs2070985 of DCN and rs1920748, rs1920751 and rs1920752 of EPYC) being polymorphic in the Chinese population under study, and these four polymorphic markers were found not to be associated with high myopia in a study of 120 cases (SE ≤ −10.0 D) and 137 controls. ³¹ Two DCN and three EPYC tSNPs were screened in the present study and were also found not to be associated with high myopia (Table 2). We did not examine rs2070985 of DCN and rs1920748 of EPYC, because their MAFs (0.089 and 0.081, respectively) documented in the HapMap database for Han Chinese are less than the selection threshold (MAF ≥ 0.10) for our study. We examined rs10859081 of EPYC (Table 2), which is in perfect linkage disequilibrium (i.e., r ² = 1; http://www.hapmap.org/) with rs1920751 and rs1920752. One Taiwanese group examined five SNPs located in either the promoter or the 3′ untranslated region of the LUM gene in 201 cases of high myopia (mean, SE ≤ −6.0 D) and 86 controls (mean SE within ±0.5 D), and found that one SNP in the 3′ untranslated region (c.1567C>T) was associated with high myopia (P = 0.0036 for allelic test and 0.0016 for genotypic test). ³² However, this SNP was not documented in the HapMap database and hence was not examined in the present study. Another Taiwanese group investigated eight SNPs in the LUM gene in 120 cases of high myopia (≤ −10.0 D) and 137 controls (−1.5 to +0.5 D) and found that rs3759223 showed a significant association with high myopia (P = 2.83 × 10⁻⁴). ³¹ Nonetheless, this association was not substantiated in two other studies of Chinese subjects. ^32,33 We did not examine rs3759223. Instead, rs2300588 was genotyped in the present study, but did not pass the initial screen by DNA pools (P = 0.4250, nested ANOVA; Table 2). Note that rs2300588 is in strong LD with rs3759223 (r ² = 0.773). This discrepancy may be due to the use of different thresholds for defining high myopia in different studies: −10.0 D in the positive study, ³¹ −8.0 D in the present study, and −6.0 D in the other two negative studies. ^32,33  

We focused on common polymorphisms of these six candidate genes, but did not explore the role of their rare variants in high myopia. A few studies did search by DNA sequence analysis for rare causal variants in the exons of FMOD or EPYC in small numbers of high myopes, but without fruitful results. ^{34 –36}  

DNA pooling cannot be used for screening X-linked candidate genes like BGN. In addition, it makes haplotype analysis extremely difficult, if not impossible. ¹⁴ However, a DNA-pooling strategy offers a very cost-effective initial screen of SNPs for follow-up studies. ¹⁴ It has also been proposed to be used in the initial phase of genome-wide association studies ³⁷ and in resequencing studies for rare variants to make the latter two approaches even more affordable. ³⁸  

In conclusion, we used an efficient three-stage approach to examining the relationship between high myopia and six candidate proteoglycan genes (ACAN, FMOD, DCN, LUM, KERA, and EPYC). In the second stage, haplotypes consisting of two ACAN SNPs (rs3784757 and rs1516794) were found to be significantly associated with high myopia. However, the initial positive result failed to be replicated in the third stage in a second subject group. Therefore, these six proteoglycan genes are unlikely to play a major role in the genetic susceptibility to high myopia. 

Supplementary Materials

Table st1, PDF - Table st1, PDF 

Table st2, PDF - Table st2, PDF 

Table st3, PDF - Table st3, PDF 

Table st4, PDF - Table st4, PDF 

Footnotes

 Supported by grants from the Research Grant Council of Hong Kong (Ref. No. PolyU 5411/06M; account code: B-Q04A) and from The Hong Kong Polytechnic University (J-BB7P, 87MS, and 87LV).

Footnotes

 Disclosure: S.P Yip, None; K.H. Leung, None; P.W. Ng, None; W.Y. Fung, None; P.C. Sham, None; M.K.H. Yap, None

The authors thank all subjects taking part in the Myopia Genetics Study.